Game Theory intro - Computer Science at UBCkevinlb/teaching/cs532a - 2006-7/lectures... · Recap:...

Recap: Utility Theory Game Theory Example Matrix Games

Game Theory intro

CPSC 532A Lecture 3

September 19, 2006

Game Theory intro CPSC 532A Lecture 3, Slide 1


Lecture Overview

Recap: Utility Theory

Game Theory

Example Matrix Games



Self-interested agents

I What does it mean to say that an agent is self-interested?I not that they want to harm other agentsI not that they only care about things that benefit themI that the agent has its own description of states of the world

that it likes, and that its actions are motivated by thisdescription

I Utility theory:I quantifies degree of preference across alternativesI understand the impact of uncertainty on these preferencesI utility function: a mapping from states of the world to real

numbers, indicating the agent’s level of happiness with thatstate of the world

I Decision-theoretic rationality: take actions to maximizeexpected utility.



Preferences Over Outcomes

If o1 and o2 are outcomesI o1 � o2 means o1 is at least as desirable as o2.

I read this as “the agent weakly prefers o1 to o2”

I o1 ∼ o2 means o1 � o2 and o2 � o1.I read this as “the agent is indifferent between o1 and o2.”

I o1 � o2 means o1 � o2 and o2 6� o1

I read this as “the agent strictly prefers o1 to o2”



Lotteries

I An agent may not know the outcomes of his actions, but mayinstead only have a probability distribution over the outcomes.

I A lottery is a probability distribution over outcomes. It iswritten

[p1 : o1, p2 : o2, . . . , pk : ok]

where the oi are outcomes and pi > 0 such that∑i

pi = 1

I The lottery specifies that outcome oi occurs with probabilitypi.

I We will consider lotteries to be outcomes.



Preference Axioms: Completeness

I Completeness: A preference relationship must be definedbetween every pair of outcomes:

∀o1∀o2 o1 � o2 or o2 � o1



Preference Axioms: Transitivity

I Transitivity: Preferences must be transitive:

if o1 � o2 and o2 � o3 then o1 � o3

I This makes good sense: otherwiseo1 � o2 and o2 � o3 and o3 � o1.

I An agent should be prepared to pay some amount to swapbetween an outcome they prefer less and an outcome theyprefer more

I Intransitive preferences mean we can construct a “moneypump”!



Preference Axioms

Monotonicity: An agent prefers a larger chance of getting a betteroutcome to a smaller chance:

I If o1 � o2 and p > q then

[p : o1, 1− p : o2] � [q : o1, 1− q : o2]



Preference Axioms

Let P`(oi) denote the probability that outcome oi is selected bylottery `. For example, if ` = [0.3 : o1; 0.7 : [0.8 : o2; 0.2 : o1]] thenP`(o1) = 0.44 and P`(o3) = 0.

Decomposability: (“no fun in gambling”). If ∀oi ∈ O,P`1(oi) = P`2(oi) then `1 ∼ `2.



Preference Axioms

I Substitutability: If o1 ∼ o2 then for all sequences of one ormore outcomes o3, . . . , ok and sets of probabilitiesp, p3, . . . , pk for which p +

∑ki=3 pi = 1,

[p : o1, p3 : o3, . . . , pk : ok] ∼ [p : o2, p3 : o3, . . . , pk : ok].



Preference Axioms

Continuity: Suppose o1 � o2 and o2 � o3, then there exists ap ∈ [0, 1] such that o2 ∼ [p : o1, 1− p : o3]



Preferences and utility functions

Theorem (von Neumann and Morgenstern, 1944)

If an agent’s preference relation satisfies the axioms Completeness,Transitivity, Decomposability, Substitutability, Monotonicity andContinuity then there exists a function u : O → [0, 1] with theproperties that:

1. u(o1) ≥ u(o2) iff the agent prefers o1 to o2; and

2. when faced about uncertainty about which outcomes he willreceive, the agent prefers outcomes that maximize theexpected value of u.

Proof idea for part 2:

I define the utility of the best outcome u(o) = 1 and of theworst u(o) = 0

I now define the utility of each other outcome o as the p forwhich o ∼ [p : o; (1− p) : o].



Lecture Overview


Game Theory




Non-Cooperative Game Theory

I What is it?

I mathematical study of interaction between rational,self-interested agents

I Why is it called non-cooperative?I while it’s most interested in situations where agents’ interests

conflict, it’s not restricted to these settingsI the key is that the individual is the basic modeling unit, and

that individuals pursue their own interestsI cooperative/coalitional game theory has teams as the central

unit, rather than agents




I What is it?I mathematical study of interaction between rational,

self-interested agents










I Why is it called non-cooperative?

I while it’s most interested in situations where agents’ interestsconflict, it’s not restricted to these settings

I the key is that the individual is the basic modeling unit, andthat individuals pursue their own interests

I cooperative/coalitional game theory has teams as the centralunit, rather than agents



TCP Backoff Game

Game Theory

Consider this situation as a two-player game:both use a correct implementation: both get 1 ms delayone correct, one defective: 4 ms delay for correct, 0 ms for defectiveboth defective: both get a 3 ms delay.

Should you send your packets using correctly-implemented TCP (which has a “backoff” mechanism) or using a defectiveimplementation (which doesn’t)?



TCP Backoff Game

Game Theory

Consider this situation as a two-player game:both use a correct implementation: both get 1 ms delayone correct, one defective: 4 ms delay for correct, 0 ms for defectiveboth defective: both get a 3 ms delay.

Should you send your packets using correctly-implemented TCP (which has a “backoff” mechanism) or using a defectiveimplementation (which doesn’t)?

Should you send your packets using correctly-implemented TCP(which has a “backoff” mechanism) or using a defectiveimplementation (which doesn’t)?

I Consider this situation as a two-player game:I both use a correct implementation: both get 1 ms delayI one correct, one defective: 4 ms delay for correct, 0 ms for

defectiveI both defective: both get a 3 ms delay.



TCP Backoff Game

I Consider this situation as a two-player game:I both use a correct implementation: both get 1 ms delayI one correct, one defective: 4 ms delay for correct, 0 ms for

defectiveI both defective: both get a 3 ms delay.

I Questions:I What action should a player of the game take?I Would all users behave the same in this scenario?I What global patterns of behaviour should the system designer

expect?I Under what changes to the delay numbers would behavior be

the same?I What effect would communication have?I Repetitions? (finite? infinite?)I Does it matter if I believe that my opponent is rational?



Defining Games

I Finite, n-person game: 〈N,A, u〉:I N is a finite set of n players, indexed by iI A = A1, . . . , An is a set of actions for each player i

I a ∈ A is an action profile

I u = {u1, . . . , un}, a utility function for each player, whereui : A 7→ R

I Writing a 2-player game as a matrix:I row player is player 1, column player is player 2I rows are actions a ∈ A1, columns are a′ ∈ A2

I cells are outcomes, written as a tuple of utility values for eachplayer



Lecture Overview


Game Theory




Games in Matrix Form

Here’s the TCP Backoff Game written as a matrix (“normalform”).

56 3 Competition and Coordination: Normal form games

when congestion occurs. You have two possible strategies: C(for using a Correctimplementation) and D (for using a Defective one). If both you and your colleagueadopt C then your average packet delay is 1ms (millisecond).If you both adopt D thedelay is 3ms, because of additional overhead at the network router. Finally, if one ofyou adopts D and the other adopts C then the D adopter will experience no delay at all,but the C adopter will experience a delay of 4ms.

These consequences are shown in Figure 3.1. Your options arethe two rows, andyour colleague’s options are the columns. In each cell, the first number representsyour payoff (or, minus your delay), and the second number represents your colleague’spayoff.1TCP user’s

game

Prisoner’sdilemma game

C D

C −1,−1 −4, 0

D 0,−4 −3,−3

Figure 3.1 The TCP user’s (aka the Prisoner’s) Dilemma.

Given these options what should you adopt, C or D? Does it depend on what youthink your colleague will do? Furthermore, from the perspective of the network opera-tor, what kind of behavior can he expect from the two users? Will any two users behavethe same when presented with this scenario? Will the behavior change if the networkoperator allows the users to communicate with each other before making a decision?Under what changes to the delays would the users’ decisions still be the same? Howwould the users behave if they have the opportunity to face this same decision with thesame counterpart multiple times? Do answers to the above questions depend on howrational the agents are and how they view each other’s rationality?

Game theory gives answers to many of these questions. It tells us that any rationaluser, when presented with this scenario once, will adopt D—regardless of what theother user does. It tells us that allowing the users to communicate beforehand willnot change the outcome. It tells us that for perfectly rational agents, the decision willremain the same even if they play multiple times; however, ifthe number of times thatthe agents will play this is infinite, or even uncertain, we may see them adopt C.

3.2 Games in normal form

The normal form, also known as thestrategicor matrix form, is the most familiargame instrategic form

game in matrixform

representation of strategic interactions in game theory.

1. The term ‘Prisoners’ Dilemma’ for this famous game theoretic situation derives from the original storyaccompanying the numbers. Imagine the players of the game are twoprisoners suspected of a crime ratherthan network users, that you each can either Confess to the crime or Deny it, and that the absolute values ofthe numbers represent the length of jail term each of you will get in each scenario.

c©Shoham and Leyton-Brown, 2006

Play this game with someone near you, repeating five times.



More General Form

Prisoner’s dilemma is any game58 3 Competition and Coordination: Normal form games

C D

C a, a b, c

D c, b d, d

Figure 3.3 Any c > a > d > b define an instance of Prisoner’s Dilemma.

To fully understand the role of the payoff numbers we would need to enter intoa discussion ofutility theory. Here, let us just mention that for most purposes, theutility theoryanalysis of any game is unchanged if the payoff numbers undergo anypositive affinepositive affine

transformation transformation; this simply means that each payoffx is replaced by a payoffax + b,wherea is a fixed positive real number andb is a fixed real number.

There are some restricted classes of normal-form games thatdeserve special men-tion. The first is the class ofcommon-payoff games. These are games in which, forevery action profile, all players have the same payoff.

Definition 3.2.2 A common payoff game, or team game, is a game in which for allcommon-payoffgame

team game

action profilesa ∈ A1 × · · · × An and any pair of agentsi, j, it is the case thatui(a) = uj(a).

Common-payoff games are also calledpure coordination games, since in such gamespure-coordinationgame

the agents have no conflicting interests; their sole challenge is to coordinate on anaction that is maximally beneficial to all.

Because of their special nature, we often represent common value games with anabbreviated form of the matrix in which we list only one payoff in each of the cells.

As an example, imagine two drivers driving towards each other in a country withouttraffic rules, and who must independently decide whether to drive on the left or on theright. If the players choose the same side (left or right) they have some high utility, andotherwise they have a low utility. The game matrix is shown inFigure 3.4.

Left Right

Left 1 0

Right 0 1

Figure 3.4 Coordination game.

At the other end of the spectrum from pure coordination gameslie zero-sum games,zero-sum gamewhich (bearing in mind the comment we made earlier about positive affine transforma-tions) are more properly calledconstant-sum games. Unlike common-payoff games,constant-sum

games


with c > a > d > b.



Games of Pure Competition

Players have exactly opposed interests

I There must be precisely two players (otherwise they can’thave exactly opposed interests)

I For all action profiles a ∈ A, u1(a) + u2(a) = c for someconstant c

I Special case: zero sum

I Thus, we only need to store a utility function for one playerI in a sense, it’s a one-player game



Matching Pennies

One player wants to match; the other wants to mismatch.

3.2 Games in normal form 59

constant-sum games are meaningful primarily in the contextof two-player (though notnecessarily two-strategy) games.

Definition 3.2.3 A normal form game isconstant sumif there exists a constantc suchthat for each strategy profilea ∈ A1 ×A2 it is the case thatu1(a) + u2(a) = c.

For convenience, when we talk of constant-sum games going forward we will alwaysassume thatc = 0, that is, that we have a zero-sum game. If common-payoff gamesrepresent situations of pure coordination, zero-sum gamesrepresent situations of purecompetition; one player’s gain must come at the expense of the other player.

As in the case of common-payoff games, we can use an abbreviated matrix form torepresent zero-sum games, in which we write only one payoff value in each cell. Thisvalue represents the payoff of player 1, and thus the negative of the payoff of player 2.Note, though, that whereas the full matrix representation is unambiguous, when we usethe abbreviation we must explicit state whether this matrixrepresents a common-payoffgame or a zero-sum one.

A classical example of a zero-sum game is the game ofmatching pennies. In this matchingpennies gamegame, each of the two players has a penny, and independently chooses to display either

heads or tails. The two players then compare their pennies. If they are the same thenplayer 1 pockets both, and otherwise player 2 pockets them. The payoff matrix isshown in Figure 3.5.

Heads Tails

Heads 1 −1

Tails −1 1

Figure 3.5 Matching Pennies game.

The popular children’s game ofRock, Paper, Scissors, also known asRochambeau, Rock, Paper,Scissors,orRochambeaugame

provides a three-strategy generalization of the matching-pennies game. The payoffmatrix of this zero-sum game is shown in Figure 3.6. In this game, each of the twoplayers can choose either Rock, Paper, or Scissors. If both players choose the sameaction, there is no winner, and the utilities are zero. Otherwise, each of the actionswins over one of the other actions, and loses to the other remaining action.

In general, games tend to include elements of both coordination and competition.Prisoner’s Dilemma does, although in a rather paradoxical way. Here is another well-known game that includes both elements. In this game, calledBattle of the Sexes, ahusband and wife wish to go to the movies, and they can select among two movies:“Violence Galore (VG)” and “Gentle Love (GL)”. They much prefer to go togetherrather than to separate movies, but while the wife prefers VGthe husband prefers GL.The payoff matrix is shown in Figure 3.7. We will return to this game shortly.

Multi Agent Systems, draft of February 11, 2006




Rock-Paper-Scissors

Generalized matching pennies.60 3 Competition and Coordination: Normal form games

Rock Paper Scissors

Rock 0 −1 1

Paper 1 0 −1

Scissors −1 1 0

Figure 3.6 Rock, Paper, Scissors game.

VG GL

VG 2, 1 0, 0

GL 0, 0 1, 2

Figure 3.7 Battle of the Sexes game.

3.2.2 Strategies in normal-form games

We have so far defined the actions available to each player in agame, but not yet hisset ofstrategies, or his available choices. Certainly one kind of strategy isto selecta single action and play it; we call such a strategy apure strategy, and we will usepure strategythe notation we have already developed for actions to represent it. There is, however,another, less obvious type of strategy; a player can choose to randomize over the set ofavailable actions according to some probability distribution; such a strategy is calleda mixed strategy. Although it may not be immediately obvious why a player shouldmixed strategyintroduce randomness into his choice of action, in fact in a multi-agent setting the roleof mixed strategies is critical. We will return to this when we discuss solution conceptsfor games in the next section.

We define a mixed strategy for a normal form game as follows.

Definition 3.2.4 Let (N, (A1, . . . , An), O, µ, u) be a normal form game, and for anysetX let Π(X) be the set of all probability distributions overX. Then the set ofmixedstrategiesfor player i is Si = Π(Ai). The set ofmixed strategy profilesis simply themixed strategy

profiles Cartesian product of the individual mixed strategy sets,S1 × · · · × Sn.

By si(ai) we denote the probability that an actionai will be played under mixedstrategysi. The subset of actions that are assigned positive probability by the mixedstrategysi is called thesupportof si.

Definition 3.2.5 Thesupportof a mixed strategysi for a player i is the set of purestrategies{ai|si(ai) > 0}.


...Believe it or not, there’s an annual international competition forthis game!



Games of Cooperation

Players have exactly the same interests.

I no conflict: all players want the same things

I ∀a ∈ A,∀i, j, ui(a) = uj(a)I we often write such games with a single payoff per cell

I why are such games “noncooperative”?



Coordination Game

Which side of the road should you drive on?

58 3 Competition and Coordination: Normal form games

C D

C a, a b, c

D c, b d, d

Figure 3.3 Any c > a > d > b define an instance of Prisoner’s Dilemma.

To fully understand the role of the payoff numbers we would need to enter intoa discussion ofutility theory. Here, let us just mention that for most purposes, theutility theoryanalysis of any game is unchanged if the payoff numbers undergo anypositive affinepositive affine

transformation transformation; this simply means that each payoffx is replaced by a payoffax + b,wherea is a fixed positive real number andb is a fixed real number.

There are some restricted classes of normal-form games thatdeserve special men-tion. The first is the class ofcommon-payoff games. These are games in which, forevery action profile, all players have the same payoff.

Definition 3.2.2 A common payoff game, or team game, is a game in which for allcommon-payoffgame

team game

action profilesa ∈ A1 × · · · × An and any pair of agentsi, j, it is the case thatui(a) = uj(a).

Common-payoff games are also calledpure coordination games, since in such gamespure-coordinationgame

the agents have no conflicting interests; their sole challenge is to coordinate on anaction that is maximally beneficial to all.

Because of their special nature, we often represent common value games with anabbreviated form of the matrix in which we list only one payoff in each of the cells.

As an example, imagine two drivers driving towards each other in a country withouttraffic rules, and who must independently decide whether to drive on the left or on theright. If the players choose the same side (left or right) they have some high utility, andotherwise they have a low utility. The game matrix is shown inFigure 3.4.

Left Right

Left 1 0

Right 0 1

Figure 3.4 Coordination game.

At the other end of the spectrum from pure coordination gameslie zero-sum games,zero-sum gamewhich (bearing in mind the comment we made earlier about positive affine transforma-tions) are more properly calledconstant-sum games. Unlike common-payoff games,constant-sum

games




Game Theory intro - Computer Science at UBCkevinlb/teaching/cs532a - 2006-7/lectures... · Recap:...

Documents

Transcript of Game Theory intro - Computer Science at UBCkevinlb/teaching/cs532a - 2006-7/lectures... · Recap:...